JP2008506176A - Method and apparatus for image processing - Google Patents

Abstract

  A method for image processing, processing an image including a plurality of pixels each having a luminance and generating an image-processed image including the plurality of pixels, applying a storage filter to the image; Using at least one stored value corresponding to at least one of the plurality of pixels of the image, calculating a weighting factor a for each of the at least one stored value, and using each of the at least one weighting factor Mixing the image and the image processed image to produce an output image that includes a plurality of pixels.

Description

CROSS REFERENCE TO RELATED APPLICATION This application claims the benefit of US Provisional Application No. 60 / 586,550, filed July 9, 2004.

  The present invention relates to an image processing method and apparatus for improving image quality and reducing noise in ultrasound images.

  Image processing typically improves image quality by smoothing, sharpening, and improving overall image contrast. For example, smoothing can reduce noise and improve contrast. However, in the case of medical imaging, smoothing can result in the loss of some clinically important features. Such important image features are generally detected as discontinuities by a discontinuity detection filter. Discontinuity detection can be performed at the edge, line, or point level. Referring to FIG. 1, a prior art edge detection filter 11 is illustrated. Referring to FIG. 2, a prior art line detection filter 21 is illustrated. Referring to FIG. 3, a conventional point detection filter 31 is illustrated. Unfortunately, the prior art filters 11, 21, and 31 are all sensitive to image noise or signal-to-noise ratio. Therefore, there is a need for a filter for both detection and storage of important image features.

  Accordingly, it is an object of the present invention to provide an image processing method and apparatus that improves image quality and reduces noise in an ultrasound image.

  According to the present invention, a method for image processing processes an image including a plurality of pixels each having a luminance, and generates an image-processed image including a plurality of pixels. Applying a conservative filter to generate at least one conserved value corresponding to at least one of the plurality of pixels of the image, calculating a weighting factor α for each of the at least one conserved value, and at least one weighting factor Are used to mix the image with the imaged image to produce an output image that includes a plurality of pixels.

  According to the present invention, an apparatus for image processing processes an image including a plurality of pixels each having a luminance, and generates an image after image processing including a plurality of pixels. An apparatus for applying a conservative filter to generate at least one conserved value corresponding to at least one of the plurality of pixels of the image, an apparatus for calculating a weighting factor α for each of the at least one conserved value, and at least An apparatus for mixing an image with an image-processed image using each of one weighting factor to produce an output image that includes a plurality of pixels is included.

  The teachings of the present invention therefore provide for any form of image processing on the image, particularly smoothing and edges that reduce spurious noise and undesirable artifacts while preserving any small features that represent the underlying actual structure. It is to provide a way to perform the emphasis. This is the statistical likelihood that any given pixel differs from its neighboring pixels for reasons related to structural differences and the composition of the region corresponding to that pixel, and not as a result of noise or error in the image. This is accomplished by applying a preservation filter to the image to determine the degree. After this conserved filter is applied to calculate a conserved value corresponding to the probability that a given pixel is the result of an actual structure, that conserved value is used and executed on that pixel. The amount of image processing, i.e., smoothing or other processing that can contribute to reducing the impact of the pixels is adjusted. In short, if the stored value indicates that a given pixel is an actual structure, the effect of image processing that can contribute to smoothing that kind of pixel on that image is reduced. . Conversely, if the stored value indicates that the pixel under examination is unlikely to represent the actual structure, but rather spurious noise or other errors injected into the image, smoothing or other Image processing that contributes to changing the value of the seed pixel is allowed to have a greater effect.

  As noted above, image processing takes many forms and is not intended to be limiting, but includes noise enhancement such as edge enhancement, contrast, and speckle.

  Referring to FIG. 5, an image filter 53 and an image 51 as illustrated in the art are illustrated. The image filter 53 is normally configured in an odd row × odd column form so that the center pixel 55 exists in the center of the image filter 53. A filter coefficient (not shown) is supplied to the intersection of each row and column of the image filter 53. Similarly, the image 51 includes a plurality of pixels arranged in rows and columns. When the image filter 53 is placed over the portion of the image 51, each pixel value of the underlying image 51 corresponding to each pixel of the image filter 53 is usually multiplied by an image filter coefficient. Once the multiplication is done, the result is added, subtracted, multiplied, or statistically manipulated to provide the image processing pixel value. This image processing pixel value is generally assigned to the pixel of the image 51 corresponding to the central pixel 55 of the image filter 53. The image filter 53 is typically applied to the image 51 by placing an image filter 53 that includes a central pixel 55 over the pixels that make up the image 51. By doing so, a processed image (not shown) corresponding to the image 51 is created, whereby the pixel value of each pixel in the processed image is compared to the image filter 53 for the pixels of the image 51. Result of the calculation performed by.

  In a preferred embodiment of the present invention, the image processing can be any type, adaptive or non-adaptive, for example a two-dimensional image filter 53 or low-pass as shown in FIG. It may include a multi-dimensional filter, a high-pass filter or band-pass filter or median filter, or a speckle reduction filter or any other type of filtering. The 5 × 5 image filter 53 of FIG. 5 is for illustrative purposes only, and may be a spatial domain filter or a frequency domain filter of other sizes.

As described above, image processing is a powerful tool for extracting a true image from a noisy image. However, image processing can sometimes result in real image loss. The present invention thus teaches a method for detecting and storing strong features in the original image. Referring to FIG. 6, the method of the present invention is illustrated. As will be described more fully later, the image is processed to produce an image Z _ip after image processing. A feature detection filter 41 is applied to the original image 51 within a small region corresponding to the size of the feature detection filter 41, moved and applied to another small region, so that a stored value for at least one pixel of the image is obtained. Is calculated. Thus, the stored value will probably be different from one region to another based on one pixel at a time. Finally, the stored value is used to adjust the amount of processed image that is blended into the original image to produce an output image on a pixel-by-pixel basis.

Still referring to FIG. 6, in step 1, the original image is image processed to produce an image processed image Z _ip . Such image processing can be of any type well known in the art and / or as described above.

Referring to step 2, a feature detection filter is applied to the original image so that a stored value is calculated for at least one pixel of the image. Referring to FIG. 4, a feature detection filter 41 having a size of (2P + 1) × (2Q + 1) pixels is illustrated. In one embodiment, when the feature detection filter 41 is applied to the image 51 (Z _ij ) to be processed, the neighboring pixels of the target pixel 57 as defined by equation (1) With the calculation of the stored value as the sum of absolute differences (SAD) in the value.

  In this case, the target pixel 57 corresponds to the coordinates (0, 0) of the feature detection filter 41. The absolute difference from the target pixel at the center or (0,0) is calculated for each pixel or each position at (i, j) in the feature detection filter illustrated in FIG. Although a 5 × 5 SAD filter (P = 2, Q = 2) is illustrated here for illustrative purposes, the size of the feature detection filter is smaller (eg, 3 × 3) or larger ( For example, it may be 7 × 7, 9 × 9, or the like. The absolute difference at each pixel is summed as expressed in equation (1). The SAD can be further normalized by the number of pixels in the feature detection filter minus 1 (corresponding to the target pixel 57) as expressed in equation (2) below.

  It is also possible to use a sum of square differences (SSD) defined as follows.

  The SSD can also be further normalized as follows.

  Although the sum of absolute differences is used for speeding up, other statistical forms (eg, variance) can be used to detect strong image features such as bright spots or points. This feature detection filter also uses image pixel values. If the sum of absolute differences is very high and the pixel value is very high or bright, it is considered that a very strong feature has been found, without applying a smoothing or speckle noise reduction filter or others Saved.

Depending on the SAD, NSAD, SSD, or NSSD value and the brightness of the target pixel, the original image Z _orig and the filtered image Z _ip are shown in Equation 5 and step 3 of FIG. As shown, they are mixed on a pixel-by-pixel basis. For example, if the brightness of the SAD and target pixel is high, then when the output image is calculated by mixing the processed image with the original image, more original images are used and less processed An image is used. The degree to which each image is included in each image when it is combined with the processed image to produce the final image output is controlled by a weighting factor α, as defined by the following equation:
Z _out = α · Z _ip + (1−α) Z _orig (5)
_Where Z _out : final image pixel output;
α: weighting factor corresponding to the target pixel;
Z _ip : pixel data after image processing;
Z _orig : Pixel data of the original image of the image 51.

  α depends on the amplitude (or luminance) I and SAD values of the original image, as described below.

In a preferred embodiment (for an 8-bit (or maximum 255) grayscale image) if I> 80,
If 1680>SAD> 720 and I> 80, α = 1− (SAD−720) / 960 (6)
If SAD> = 1680 and I> 80, α = 0
If SAD <= 720 and I> 80, α = 1.0
For I <80, for a 5 × 5 filter for an 8-bit grayscale image with a maximum luminance of 255 (or a range of 0-255), α = 1.0 regardless of SAD. Become. In the preferred embodiment, α is quickly obtained using the SAD and I look-up table (LUT) rather than the actual computation of equation (6), which takes more time. The above formulas and expressions merely illustrate the relationship between SAD, I, and α.

  Although described with reference to grayscale image luminance values and corresponding SAD / NSAD / SSD / NSSD values, these are also referred to as conservative statistics measures, for both I and SAD / NSAD / SSD / NSSD values. Is shown as the absolute range of luminance and SAD / NSAD / SSD / NSSD values. For example, a value of “I> 80” is a brightness greater than 80 at 255, the maximum possible brightness for an 8-bit grayscale image. Similarly, for an image having a luminance range for each pixel ranging from zero to maximum luminance, the SAD / NSAD / SSD / NSSD values range from zero to the maximum value. Thus, “1680> SAD> 720” refers to the SAD value between 720 and 1680 on the absolute scale.

Since the image can have a luminance resolution different from 8 bits, the above relationship can be more generally represented by SAD and I normalized by their respective maximum values. These values are called SAD% and I%. α can be expressed by the following equation.
If 27.5%>SAD%> 11.8% and I%> 31.4%,
α = 1− (SAD% −11.8%) / 15.7% (7)
If SAD%> = 27.5% and I%> 31.4%,
α = 0
If SAD% <= 11.8% and I%> 31.4%,
α = 1.0
If I% <31.4%, then α = 1.0 for all SAD%.

  Although described with reference to grayscale image luminance values and corresponding SAD / NSAD / SSD / NSSD values, these are also referred to as conservative statistics measures, for both I and SAD / NSAD / SSD / NSSD values. Can be shown as a normalized range of luminance and SAD / NSAD / SSD / NSSD values, for example as a percentage of the range from zero to 100. For example, “I%> 31.4%” has a luminance greater than 31.4% of the maximum possible luminance regardless of the luminance of the pixel defined. Similarly, for an image having a luminance range for each pixel ranging from zero to maximum luminance, the SAD / NSAD / SSD / NSSD values range from zero to the maximum value. Therefore, “27.5%> SAD%> 11.8%” refers to a SAD% value between 11.8% and 27.5% of the maximum SAD / NSAD / SSD / NSSD value.

  Although the above embodiment uses exemplary boundary values for both SAD / NSAD / SSD / NSSD values and luminance values, such boundary values are exemplary and vary depending on the image creation situation. May be. Such boundary values are preferably selected so that small structures of interest in the image are preserved while sufficient to allow substantial elimination of the adverse effects of noise or other data.

  Regardless of the boundary value selected, α is assigned according to the following method. If the brightness is less than the boundary value (eg 80), then α is equal to the maximum α value (preferably ˜1.0) regardless of the SAD / NSAD / SSD / NSSD value or the value of the stored statistical measure. Become. If the luminance exceeds the boundary value, α is equal to the maximum α value if the SAD / NSAD / SSD / NSSD value is within a low region (eg, SAD <= 720 or SAD% <= 11.8%). If the SAD / NSAD / SSD / NSSD value is in the upper region (for example, SAD> = 1680, or SAD%> = 27.5%), α becomes the minimum α value (preferably to 0.0). In the middle region (eg, 1680> SAD> 720, or 27.5%> SAD%> 11.8%), α is linearly correlated to the SAD / NSAD / SSD / NSSD value.

  As a result, in the case of the above example, if the image brightness I is above the 80 brightness threshold (or if I% is above 31.4%), then in the middle region (or SAD is 720) Α is linearly proportional to the SAD / NSAD / SSD / NSSD value between 1680 and between 11.8% and 27.5%). As described above, this type of value is described herein in an exemplary manner and depends on the type of image and is different in one application (eg, liver) and another (eg, fetus). In this case, 80 image intensities were used as thresholds, but this is only for illustrative simplicity of the method. In another embodiment, the 80 thresholds can be changed depending on the SAD. In another embodiment, α can be decreased more rapidly than the linear function with SAD. It is also possible for α to be another function of SAD and I. Regardless of the relationship between I, SAD, and α, α is determined by a SAD and I two-dimensional table or look-up table (LUT). The above equation for α is for illustrative purposes only and can take any form using these two parameters (SAD / NSAD / SSD / NSSD and I). In another embodiment, SAD / NSAD / SSD / NSSD and I can be used in addition to mean and variance (standard deviation), or other statistics.

Referring to FIG. 7, a preferred embodiment of the apparatus used to perform the method of the present invention is illustrated. The processor 71 receives the original image Z _orig and outputs an output image Z _out . The processor 71 is preferably a digital signal processor. In any case, the processor 71 receives digital information formatted to represent the input image Z _orig . The processor 71 operates on the input image Z _orig and applies a storage filter, processes the original input image to produce an image processed image Z _ip, and weights each of the at least one calculated storage value α is calculated, and the original image and the image processed image are mixed to generate an output image Z _out .

  In another embodiment, the SSD can be used with another statistical value such as the mean, variance, standard deviation of the pixels in the image corresponding to the storage filter, or any other statistical measure. is there. α is also determined on a pixel-by-pixel basis and is therefore different from one point to another in the image.

  In the foregoing, one or more embodiments of the present invention have been described. However, it will be understood that various modifications may be made without departing from the spirit and scope of the invention. Accordingly, other embodiments are within the scope of the claims.

It is explanatory drawing which showed the edge detection filter (prior art). It is explanatory drawing which showed the edge detection filter (prior art). It is explanatory drawing which showed the edge detection filter (prior art). It is explanatory drawing which showed the image feature preservation | save filter. It is explanatory drawing which showed the image filter of this invention. It is explanatory drawing which showed the image filter of this invention. 3 is a flowchart of the method of the present invention. 1 is a schematic view of an apparatus of the present invention.

Claims (9)

Processing an image including a plurality of pixels, each having a luminance, and generating an image processed image including the plurality of pixels;
Applying a storage filter to the image to generate at least one stored value corresponding to at least one of the plurality of pixels of the image;
Calculating a weighting factor α for each of the at least one stored value;
Mixing the image and the image processed image using each of the at least one weighting factor to produce an output image comprising a plurality of pixels;
A method for image processing comprising: Processing the image is from the group consisting of smoothing, edge enhancement, low-pass filtering, high-pass filtering, band-pass filtering, median filtering, speckle reduction filter, noise reduction, and contrast enhancement. Including performing selected image processing,
The method according to claim 1. The application of the preservation filter is
Applying the storage filter to the image to calculate at least one storage statistical measure each corresponding to a single pixel of the image;
The method according to claim 1. The application of the preservation filter includes sum of absolute differences (SAD), normalized sum of absolute differences (NSAD), sum of square differences (SSD), normalized sum of square differences (NSSD), average, Calculating a statistical measure of the at least one stored selected from the group consisting of variance and standard deviation;
The method according to claim 3. The calculation of the weighting factor includes calculating the weighting factor from the luminance of the plurality of pixels of the image corresponding to the storage filter and the at least one stored value.
The method according to claim 1. The calculation of the weighting factor comprises using a look-up table (LUT);
6. The method of claim 5, wherein: The calculation of the weighting factor α is
Setting α to a maximum value when the brightness of the target pixel is less than or equal to the boundary value;
Setting α to a maximum value when the brightness of the target pixel is greater than the boundary value and the stored value corresponding to the target pixel is in a low region;
Setting α to a minimum value when the brightness of the target pixel is greater than the boundary value and the stored value corresponding to the target pixel is in the upper region;
Setting α in proportion to a stored statistical measure when the brightness of the target pixel is greater than the boundary value and the stored value corresponding to the target pixel is in an intermediate region;
The method of claim 5 comprising: Generation of the output _image, when the corresponding pixel of the image pixels, and a _{Z orig} in the image in the image after processing the pixels of the _{Z out} the output _image, the _{Z ip,} wherein _{Z out} = alpha Calculating each of the plurality of pixels of the output image according to Z _ip + (1−α) Z _orig ;
The method according to claim 7. Means for processing an image including a plurality of pixels, each having a luminance, and generating an image processed image including the plurality of pixels;
Means for applying a storage filter to the image to generate at least one stored value corresponding to at least one of the plurality of pixels of the image;
Means for calculating a weighting factor α for each of the at least one stored value;
Means for mixing the image and the imaged image using each of the at least one weighting factor to produce an output image comprising a plurality of pixels;
An apparatus for image processing, comprising:

Images